Interference with correlated photons: Five quantum mechanics experiments for

undergraduates
E. J. Galvez, Charles H. Holbrow, M. J. Pysher, J. W. Martin, N. Courtemanche, L. Heilig, and J. Spencer
Citation: American Journal of Physics 73, 127 (2005); doi: 10.1119/1.1796811
View online: http://dx.doi.org/10.1119/1.1796811
View Table of Contents: http://scitation.aip.org/content/aapt/journal/ajp/73/2?ver=pdfcov
Published by the American Association of Physics Teachers
Articles you may be interested in
Quantum correlation of path-entangled two-photon states in waveguide arrays with defects
AIP Advances 4, 047117 (2014); 10.1063/1.4871401
Fizeaus aether-drag experiment in the undergraduate laboratory
Am. J. Phys. 80, 497 (2012); 10.1119/1.3690117
Low-cost coincidence-counting electronics for undergraduate quantum optics
Am. J. Phys. 77, 667 (2009); 10.1119/1.3116803
Photolithographic fabrication of diffraction and interference slit patterns for the undergraduate laboratory
Am. J. Phys. 72, 1328 (2004); 10.1119/1.1775240
Observing the quantum behavior of light in an undergraduate laboratory
Am. J. Phys. 72, 1210 (2004); 10.1119/1.1737397

This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

Interference with correlated photons: Five quantum mechanics experiments

for undergraduates
E. J. Galvez, C. H. Holbrow, M. J. Pysher, J. W. Martin, N. Courtemanche,
L. Heilig, and J. Spencer
Department of Physics and Astronomy, Colgate University, Hamilton, New York 13346

Received 15 March 2004; accepted 29 July 2004

We describe five quantum mechanics experiments that have been designed for an undergraduate
setting. The experiments use correlated photons produced by parametric down conversion to
generate interference patterns in interferometers. The photons are counted individually. The
experimental results illustrate the consequences of multiple paths, indistinguishability, and
entanglement. We analyze the results quantitatively using plane-wave probability amplitudes
combined according to Feynmans rules, the state-vector formalism, and amplitude packets. The
apparatus fits on a 2 4 optical breadboard. 2005 American Association of Physics Teachers.
DOI: 10.1119/1.1796811

I. INTRODUCTION
Advances in laboratory techniques for doing experiments
with single photons have stimulated studies of the fundamentals of quantum mechanics that underlie such interesting applications as quantum cryptography and quantum computing.1 In particular, the ability to produce pairs of correlated photons allows us to bring beautiful laboratory demonstrations of quantum superposition to an undergraduate setting where simplicity and affordability are primary
concerns.2
In this article we describe five table-top experiments that
involve the interference of photons detected by a counting
apparatus. The experiments involve photons passing through
an interferometer, where alternative paths can be made distinguishable or indistinguishable. These experiments can
provide the basis for an undergraduate laboratory on the fundamentals of quantum mechanics as proposed in Ref. 3.
They go beyond transforming interferometer fringes into
counter clicks and challenge classical intuition with results
that are unquestionably nonclassical. By incorporating these
experiments into undergraduate quantum mechanics instruction, we hope to encourage students to discuss and consider
the consequences of quantum mechanical superposition such
as entanglement and nonlocality.
The experiments have the attractive feature that their results can be analyzed and understood by undergraduates. We
try to explain them in ways that we believe will be useful
and accessible to them. Our explanations assume that they
are acquainted with the basic ideas of interference and wave
packets and that they have learned, or can quickly learn, to
use the complex exponential representation of plane waves
what Feynman ingeniously described as clock numbers. 4
The first of the five experiments demonstrates that a photon interferes with itself when it can reach a detector by
either of two indistinguishable paths. We observe this effect
by changing the phase of one of the paths without making
the paths distinguishable. We also show what happens to the
interference pattern when, using filters and other optical elements, we modify the extent to which the two paths are
indistinguishable. In the second experiment we pass an entangled pair of photons through an interferometer and observe and analyze the unusual interference properties of this
biphoton. In the third experiment, we create photons in po-

larization states and manipulate these states as examples of

the formation, projection, and transformation of quantum
states. We also cause the interference pattern to disappear by
manipulating the polarization states of the photons to make
the paths through the interferometer distinguishable. A fourth
experiment, the quantum eraser, demonstrates how interference can be made manifest in subsets of events that together
exhibit no interference. Finally, we perform a conceptually
simple experiment that can show that the photon does not
split.
Most of our experiments and layouts are based on published landmark experiments on the fundamentals of quantum mechanics. Our references show the sources that we
consulted, but they are not chronological or comprehensive.
The cost of the experiments ranges from $14,000 to
$35,000 depending on the equipment at hand. The cost is
dominated by the price of a blue laser $2000$6000 and
two avalanche photodiode detectors $4000 each. These
prices are likely to decrease in the near future as the technologies mature. The cost of the remaining items depends on
the availability of optical hardware and conventional electronics. In Appendix A we list vendors and the prices of the
components.
II. APPARATUS
A. The laser
At the heart of the experiments is the production of a pair
of photons by spontaneous parametric down conversion, a
nonlinear effect that produces two photons from one pump
photon.5 For historical reasons the two outgoing photons are
usually called the idler and the signal photon. We use
the subscripts p, i, and s to refer to the pump, idler, and
signal photons, respectively. Conservation of energy requires
that
E p E s E i ,

where E p is the energy of the pump photon, and E s and E i

are the energies of the down-conversion photons. It often is
convenient to use the alternative forms of Eq. 1,

p i s ,

2a

k p k i k s ,

2b

127
Am. J. Phys. 73 2, February 2005
http://aapt.org/ajp
2005 American Association of Physics Teachers
127
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

1
1
1
,
p i s

2c

which are based on the fact that E kchc/ for a

photon, where is the angular frequency of the light and k is
the magnitude of its wave number in vacuum. The wave
number k is related to the wavelength in vacuum by k
2 / /c, where c is the speed of light.
High detection efficiency of down-conversion photons is
essential for our experiments and severely limits the choice
of pump laser. Efficiency is important because our experiments depend on selecting down-conversion pairs of photons
from the background. We select the photon pairs by detecting
them in coincidence, making use of the fact that two downconversion photons always are produced at nearly the same
time. Because the efficiency of coincidence detection is the
product of the individual detector efficiencies, it is necessary
to use the most efficient single-photon detectors available.
These are avalanche photodiodes. Their efficiencies peak at
around 80% for 700 nm photons and drop off rather quickly
to below 10% for 1000 nm photons.6 Because detection efficiencies vary strongly with wavelength, we concentrate on
the case where k s k i k p /2.
To operate detectors at their peak efficiency, we want
down-conversion photons with wavelengths of about 700
nm. To obtain these photons, the best pump laser is an expensive UV argon-ion laser operated at 351.1 nm, which is
widely used for research. For this wavelength there currently
are no less expensive alternatives. We used two compromises: an 18 mW, 402.36 nm GaN diode laser and a 100
mW, 457.9 nm argon-ion laser. The GaN laser has become
the choice of the compact-disk industry for data storage at
higher resolution, so its price is likely to go down.7 Avalanche photodiodes have an efficiency of about 60% at
804.72 nm. The 457.9 nm line was the shortest wavelength
available to us from an old multiline argon-ion laser that had
been used for pumping a cw dye laser. The avalanche photodiodes efficiency for the 915.8 nm down-conversion photons is 30%.
B. The crystal
The argon-ion laser produces a near-IR background glow
that needs to be removed, so it is common to use either a
dispersing prism or a dielectric mirror between the pump and
the down-converter crystal. We used both in the two experimental setups reported here. In both cases we arranged for
the polarization of the pump beam to be horizontal. The output of the argon laser was vertically polarized, so we rotated
it using mirrors.8 Standard laser safety precautions should be
used when steering the laser into the crystal.
We used nonlinear crystals cut for type-I parametric downconversion to produce a pair of down-conversion photons
with linear polarizations parallel to each other but orthogonal
to the polarization of the pump beam see Appendix B. We
set up the beta-barium-borate down-conversion crystals 57
mm thick in the arrangements shown schematically in Fig.
1. In Fig. 1a the pairs of down-conversion photons leave
the crystal at 3 to the pump-beam axis; for the arrangement in Fig. 1b they left at 0 that is, they were collinear.
In addition to energy conservation see Eq. 1, parametric
down-conversion requires that the photon momentum p is
conserved inside the crystal. The momentum is related to the
wave vector by the relation pk. The wave number inside

Fig. 1. In a the down-conversion photons travel separate paths; in b they

travel the same paths. The optical components are polarization rotator PR,
prism Pr, down-converter crystal C, nonpolarizing 50-50 beam splitter
B, half-wave plate H, polarizer P, mirror moved by a piezoelectric
stack Mp, lens L, bandpass filter F, and avalanche photodiode APD.

the crystal is k nk2 n/, with n being the index of

refraction of the crystal. This momentum conservation condition can be expressed in terms of the wave vectors as
kp ks ki .

The directions taken by the down-conversion photons of

specific but complementary wavelengths are determined by
the angle formed by the optic axis of the crystal OA and the
propagation direction of the pump beam, the phase-matching
angle m . As described in Appendix B, the down conversion
of 457.9 nm into 915.8 nm at 3 requires m26.13.
Similarly, m29.01 is required for down conversion of
402.36 nm into 804.72 nm at 0. The crystals were mounted
on a rotation stage so that OA was in a horizontal plane. In
this way we could easily fine tune the phase-matching angle
of the crystal.
C. The detectors
We chose avalanche photodiodes modules optimized for
high quantum efficiency and configured to output TTL
pulses, suitable for the electronics that we had available.
In the setup of Fig. 1a one down-conversion photon, the
idler, was sent directly to a detector, while the other photon,
the signal, was sent through a MachZehnder interferometer
to a second detector. Only coincidences of the TTL output of
both detectors were recorded. In effect the idler tags the signal photon, but in reality the photons are much more intimately connected, because their wave functions are entangled as we will discuss. In the setup of Fig. 1b there is
no distinction between idler and signal, because both photons
can travel the same path. In this case we place a beam splitter

128
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
128
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

at the interferometers output and look for coincidences between photons emerging from separate outputs of this beam
splitter.
It is very important to keep the avalanche photodiodes
from receiving ambient light, because they can be destroyed
by excess photons. To avoid doing the experiments in complete darkness, we put the detector pair of each setup inside
a light-tight box made of an aluminum frame and black
poster-board walls. The down-conversion photons enter the
detector boxes through windows covered by red filters. The
protective boxes allowed us to illuminate the work area with
low-level illumination from blue LEDs. Each avalanche photodiode had a short focal length lens 40100 mm to focus
the light onto the small active area of the avalanche photodiode 0.175 mm diameter. A narrow-band filter 10 nm, 1
nm, or 0.1 nm prevented the avalanche photodiodes from
being overwhelmed by photons that were not in the wavelength region of interest. The filters also served to determine
the bandwidth of the detected light, as described in Sec. III.
D. The electronics
We used NIM electronics nuclear instrument modules to
perform coincidence detection of the TTL pulses from the
avalanche photodiodes. The minimum electronics required
are counters and coincidence modules. Because the signal
and idler photons traveled different distances to reach the
detectors, we used a time-to-amplitude converter TAC, a
single channel analyzer SCA, and a multichannel scaler to
register coincident events. Manipulating the beams so that
both photons travel the same distance would reduce the
amount of electronics required,2 but it would have unacceptably increased the constraints on our optical layout.
We used a combined TAC/SCA unit Canberra model
2145, which had both TAC and SCA outputs. The pulses
from the signal detector passed through an extra three meters
of cable deliberately inserted to produce a 15 ns delay between the pulses produced by the detection of the signal and
idler photons. The idler and the signal pulses were sent to the
start and stop inputs of the TAC/SCA, respectively. The TAC
output was sent to the multichannel scaler for pulse-height
analysis. The output of the multichannel scaler consisted of a
histogram of the number of pairs of pulses as a function of
the time delay between them. We used three counters, two to
register the singles counts from each detector and one to
record the SCA output, that is, the coincidences. After a year
of operation, we automated the data acquisition using
Labview9 and shortened the data acquisition time from hours
to minutes.

ranged to be displaced transversely in order to fine-steer the

focused light onto the small photodetector area.
These techniques are very useful to obtain the first signals.
We start by maximizing the singles counts on each detector
expecting 10100 kHz by varying the position of the lens
in front of each detector. Then we set the TAC/SCA to record
photon pairs delayed by a time in the range 050 ns and look
for a peak in the multichannel scaler output around a 15 ns
delay. When the optical layout is aligned properly, there is a
prominent peak at that delay representing the downconversion pair. Once the signal is obtained, we narrow the
TAC/SCA window to about 4 ns centered around the downconversion peak. The SCA output then represents downconversion events. If the TAC and SCA are separate modules, the voltage window of the SCA determines the time
window.
F. Interferometers
We used both MachZehnder and Michelson interferometers. The former is more elegant but requires more optical
elements and greater care in setting up. To align it we started
with a pilot beam from a HeNe laser and put each of the
interferometer components that is, mirrors and beam splitters into place one by one. We used irises to align the beams
parallel to the holes of the breadboard.9 In the arrangement
of Fig. 1a we set up the interferometer so that we could
optimize the down-conversion and the interferometry separately. Because stability is important, we linked all the
mounting posts in the interferometers to each other by 1/2
diameter rods Invar or stainless steel. We also were able to
find adequate stability by using short pedestal mounts for the
optics instead of posts with rod links. We mounted one of the
mirrors on a translation stage to be able to adjust the difference between the path lengths of the interferometers arms to
near zero, that is, the point where white-light fringes are
observed.9 This adjustment was needed because the coherence of the down-converted light is limited. Indeed, adjusting the stage to increase the path difference to somewhat less
than 1 mm caused the fringes to disappear.
Our experiments were done by slightly varying the difference between the lengths of the arms of the interferometer by
moving one of the interferometer mirrors with a stack of
piezoelectric transducers either glued directly to the mirror
and mirror mount or placed as a spacer in the translation
stage where the mirror was mounted. The piezo had a response of about 40 nm/V. When our setup was automated,
the output of a digital-to-analog interface was fed to a highvoltage amplifier to scan the voltage on the piezo.
III. ONE-PHOTON INTERFERENCE

E. First steps
Our experimental setups were designed to take into account that the down-conversion beams are too weak to be
seen. Down-conversion efficiencies at the wavelength of interest are typically about 1010. Therefore, it is necessary to
prealign the optical components and detectors, and, of
course, any interferometer. Prealignment requires a very methodical approach: calculating the positions of the detectors,
setting up irises, and tracing the expected path of the downconverted beam with the beam from a HeNe laser. The other
requirement is flexibility: the crystal is mounted on a rotation
stage that permits fine tuning of the phase-matching angle,
and the lens in front of each avalanche photodiode is ar-

The probability amplitude, a complex number, is a key

idea in quantum mechanics. Interference arises from squaring the sum of the probability amplitudes for alternative
ways to the same observational outcome. Interference can
occur if two or more different ways to produce the same
result cannot be distinguished with the apparatus. If the apparatus yields information that can distinguish between alternatives, interference will not occur.
Feynman states the conditions for interference in three
rules:10 1 the probability P of a particular outcome from
the interaction of a particle with an apparatus is given by the
square of the absolute value of a complex probability amplitude : P 2 ; 2 when the same outcome can occur in

129
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
129
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

s enters the MachZehnder interferometer and interacts

with the first beam splitter. The beam splitter transforms the
state of the input photon into
s t 1 r 2 ,

Fig. 2. Schematic of the experiments using a MachZehnder interferometer.

The labeled components shown are source of down converted photons S,
nonpolarizing 50-50 beam splitter B, mirror moved by a piezoelectric
stack Mp, lens L, bandpass filter F, and avalanche photodiode. The
interferometer arms have lengths 1 and 2 , and its output ports are labeled
a and b.

indistinguishable alternative ways, the probability amplitude

is the sum of the probability amplitudes for each way considered separately: P 1 2 2 ; 3 when an experiment
is performed that is capable of determining which way the
outcome occurred, the probability of the outcome is the sum
of the probabilities of each alternative: P P 1 P 2 1 2
2 2.
Although Feynmans approach describes how basic interference patterns arise, a more complete theory is needed to
explain partial interference and to give students insight into
the coincidence detection of photons.
A. One-photon interference pattern: Prediction
Feynmans approach can predict the interference pattern
produced by an interferometer with nearly equal arm lengths
because the resulting interference pattern is insensitive to the
details of the amplitudes.11 We can assume the amplitudes
are plane waves modified by reflection and transmission at
the beam splitter as the photons pass through the setup of
Fig. 2. At the first beam splitter a photon has two possible
outcomes: either transmission with an amplitude t into arm 1
or reflection with amplitude r into arm 2. The plane wave
associated with path j1,2 acquires a phase j k s j ,
where j is the length of arm j and k s 2 / s is the wave
number of the signal photon.12 Therefore the probability amplitude for emerging from path 1 is tre i 1 and that from path
2 is rte i 2 .
The probability P that a photon will be detected at the a
port of the second beam splitter see Fig. 2 is the square of
the modulus of the sum of these two amplitudes:
P tre i 1 rte i 2 2

4a

rr * tt * 2e i( 1 2 ) e i( 1 2 )

4b

2RT 1cos ,

4c

where Rrr * and Ttt * are the reflection and transmission

probabilities, respectively, and 1 2 is the phase difference arising from the difference between lengths of the two
arms of the interferometer. We assume the beam splitters are
identical and symmetric.3 For the common case of a 50-50
beam splitter, R1/2 and T1/2, and Eq. 4c becomes P
1/2(1cos ). This result implies that varying will produce a variation in the number of photons emerging from the
interferometer, that is, there will be an interference pattern.
To introduce bra and ket notation for state vectors, we
could use the slightly more sophisticated explanation offered
by Greenberger, Horne, and Zeilinger.13 A photon in a state

where 1 and 2 refer to the states of the photon in the

interferometer arms. If a and b denote the state of the
signal photon exiting the a and b output ports of the interferometer, respectively, then
1 e i 1 r a t b ),
2 e

i2

t a r b ).

6a
6b

We combine Eqs. 56b and express the overall transformation of s as

s rt e i 1 e i 2 a tte i 1 rre i 2 b .

The interference pattern arises from the probability of detecting a photon in state a,
P a s 2 2RT 1cos .

Equation 8 is the same result as Eq. 4c with the desirable

feature that the formalism conveniently yields the amplitude
for a photon to arrive at the other output, that is, in state b.14
B. One-photon interference: Experimental results
To be sure we were measuring an interference pattern produced by individual photons, we used coincidences between
down-conversion pairs. We recorded the detection of a signal
photon at output a of the interferometer only when an idler
photon was detected at another detector. Thus each downconversion signal photon passing through the interferometer
was tagged by its idler companion. Such a tagging procedure
does not change the expected interference pattern, because
the crystal down converts a pump photon p into a pair of
states p i s , where , the amplitude for down conversion into a signal and an idler photon, s and i, is the
order of 105 or 106 . The effect is to multiply all the terms
in Eq. 7 by i, which does not introduce any additional
relative phase shift. As a result, the overall probability remains proportional to Eq. 8 although the count rate will
drop by 1010 or so.15
The interferometer was set up to make its arm lengths as
closely equal as possible, that is, 1 2 0, by aligning it to produce white-light fringes. Then with 10 nm bandpass filters centered on the down-conversion wavelength in
front of both detectors, we obtained the interference fringes
in the coincidence counts N c shown in Fig. 3a. The data are
the number of recorded coincidences as a function of ,
which was varied by changing 2 via the piezo-driven mirror
in the interferometer arm Mp in Fig. 1. The error bars are
proportional to N, due to Poisson statistics. The slight horizontal deformation of the sinusoid is a reflection of the nonlinear relation between the displacement of the piezo and the
voltage applied to it.
The data were corrected for accidental coincidences.
These arise from the accidental simultaneous arrival of two
unrelated photons. The number of these events is estimated
from the relation
N accN s N i T,

where N s and N i are the singles counts at the signal and the
idler detectors respectively, and T is the width of the coin-

130
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
130
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

comes zero, that is, interference disappears when the experiment is capable of distinguishing which path a photon takes
in passing through the apparatus. Our data give V0.85
0.05. 16
Note the strength of the evidence that the interference is
occurring one photon at a time. For one thing, it is very
improbable that at any given instant there is more than one
photon in the apparatus. If we take into consideration the
effect of the optics, filters, and detector, we estimate the detection efficiency to be det0.10. With the length of the
arms of the interferometer at 21.5 cm, and for a maximum recorded counts of the signal singles (N s
87500 s1 ), the average number of photons in the interferometer at any given time is never more than N s / detc
6104 . Moreover, we record only those photons detected
in coincidence with their down-conversion partner photon,
an even smaller number. These results drive home to students
that the interference pattern arises from the interference of
each individual photon with itself.17
C. One-photon interference: Entanglement,
distinguishability, and coherence
Fig. 3. Coincidence counts as a function of the voltage on the piezoelectric
stack used to change the interferometer path-length difference . The data
in a and b correspond to 0 and 144 m, respectively. For the
circles both detectors had 10 nm bandpass filters in front of them. The
squares in b correspond to having a 10 nm filter in front of the signal
detector and a 0.1 nm filter in front of the idler detector.

cidence window set by the TAC/SCA. The time window was

T4 ns. For the experiment of Fig. 3a, N i
93,000 s1 , while at the maxima of the interference pattern the signal count rate was about N s 87,500 s1 ; at the
minima it was about N s 11,800 s1 . The corresponding
maximum and minimum rates of accidental coincidences
were N acc32 s1 and N acc4 s1 .
A glance at Fig. 3 suggests that the simple theory does a
good job. To test it more closely, we fit the data with a
parameterized version of Eq. 8
N c N 0 1V cos ,

10

where ( f 0 f 1 v p ) v p 0 , with the fitting parameters

N 0 , V, f 0 , f 1 , and 0 . The quantity N 0 normalizes Eq. 8
to the data and the quantity V sets the depth of the minima.
The inclusion of the linear and quadratic frequency terms,
respectively f 0 and f 1 , in the fit takes into account the nonlinear variation of the displacement of the piezo as a function
of the applied voltage.
The quantity V is an important parameter often called the
visibility. It is defined as
V

N maxN min P max P min

,
N maxN min P max P min

11

where N max and N min are the maximum and minimum counts
in the interference oscillation. As Eq. 11 indicates, the visibility can be defined equivalently in terms of the maximum
and minimum probabilities P max and P min .
The visibility provides a measure of the completeness of
the interference; V1 represents full interference when
0. When is increased, the visibility of the interference
fringes decreases. As explained in the following, V be-

Although Feynmans approach explains simply how the

basic interference patterns arise, more complicated probability amplitudes are needed to explain partial interference and
to give students some insight into the coincidence process.
More complete probability amplitudes would include the effects of features of the apparatus such as filters, apertures,
mirrors, beam splitters, and the distances between components, as well as the spectral purity, intensity, polarization,
and angular spread of photon beams passing through the apparatus. These calculations, which can be complicated and
difficult, usually result in one or more mathematical objects
that have a spatial and temporal extent and which can be
thought of as amplitude packets analogous to the wave packets of classical theory. Interference then arises from linear
combinations of two or more overlapping packets.
As explained in the following, the basic features of an
amplitude packet are ultimately determined by the Heisenberg uncertainty principle. We can sometimes use this principle to determine the spatial or temporal spread of a packet
without doing detailed calculations. Such information makes
it possible to tell when two amplitude packets will have appreciable values at the same place at the same time. Only
then can there be interference. We can understand partial
interference as occurring from the partial overlap of amplitude packets with well defined phases.
The overlap of amplitude packets also is related to the
issue of distinguishability versus indistinguishability. When
amplitude packets corresponding to different ways for an apparatus to give rise to the same observational outcome overlap in space and time, the apparatus does not distinguish
between the different ways to the same outcome. When the
packets do not overlap, the output from the apparatus contains information distinguishing between the alternatives. Although there are some pitfalls, in general, the nonoverlap of
packets is equivalent to the distinguishability of alternatives,
and in either case there will be no interference.
The down-conversion process itself produces amplitude
packets with spatial spread. Although the pump-laser field
has a very small spread of wave numbers and so can be well
represented by plane-wave amplitudes, down-conversion
photons emerge with a considerable spread of wave numbers

131
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
131
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

k. The uncertainty principle tells us that there will be a

corresponding spatial spread of at least the order of x
1/k. Thus a large spread in k means a narrow spread in x
which in turn means that it will be technically difficult to
overlap two such amplitude packets.
The spread of the amplitude packets of the downconversion photons occurs even though conservation of energy assures that k i k s k p to within the very narrow energy spread of the pump laser. There can be appreciable
variation in the individual values of k i and k s without violating the conservation of energy, because k i is allowed to deviate from k 0 by some amount, call it k, if k s deviates by
a compensating amount k. Down conversion produces
pairs with k i k 0 k and k s k 0 k with appreciable values
for k, where k 0 is the central wave number that is, k 0
k p /2). As a result, down-conversion photon pairs have a
much wider spread of energy than the pump photons, and
they also are strongly correlated in energy because whatever
the sign and amount of k for the idler photon, the signal
photon must have k of the same magnitude and opposite
sign.
This energy correlation of the down-conversion photons is
a manifestation of an unique quantum property called entanglement. Entanglement refers to the production of two
or more particles in a state that has correlated properties
such as k 0 k and k 0 k), but where neither particle possesses a definite property until a measurement is made. In
our case entanglement means that neither down-conversion
photon is in the state k 0 k until a measurement is made.
The measurement of one photon puts it randomly into a definite state and the other photon goes automatically into the
correlated state. Thus if a measurement of the idler photon
yields k 0 k, the signal photon will then be in the state k 0
k, and vice versa. An example of a state vector representing such entanglement before the measurement is

1
&

k0 k s k0 k i k0 k s k0 k i ),

12

where the subscripts s and i refer to the signal and idler

photons. Although Eq. 12 is a special case, it illustrates the
characteristic property of an entangled state: It is a nonfactorable combination of single-particle states.
A very general representation of the possible states of a
pair of down-conversion photons is the continuous superposition of state vectors5,18,19
s,i

dk s

of the beams and by bandpass filters in front of the detectors.

By knowing the properties of these components, we can estimate the size of an amplitude packet, relate it to the coherence length, and use it to explain when an interference pattern will occur and when it will fade away.
We already mentioned that an equivalent analysis is possible in terms of distinguishability and indistinguishability.
Although indistinguishability is a somewhat slippery concept, it is reasonably clear how to apply it to our interference
experiments. By measuring the time delay between the idler
and signal photons, our apparatus can, in principle, distinguish a longer path taken by a photon passing through the
interferometer from a shorter path. The time delay between
the detection of the tag photon and the photon passing
through the interferometer will be longer when the photon
takes the longer interferometer path than when it takes the
shorter one. The time delay will be the difference in path
lengths divided by c, the speed of light.
The spread k of the wave numbers of the downconversion photons leads to a spread t in the difference
between the times when the two photons arrive at their detectors. As long as the difference in path lengths through the
two interferometer arms does not correspond to a difference
of travel time greater than t, we cannot distinguish which
path the signal photon took through the interferometer. But
when the difference in interferometer path lengths becomes large enough so that /ct, the time at which a
photon taking the longer path is detected in coincidence with
the idler will be distinguishably longer than that of a photon
taking the shorter path. In principle, we will then be able to
distinguish which path a photon takes, and there should be
no interference. The distance ct is a measure of the spatial
extent of the amplitude packet and is called its coherence
length c . Thus we expect that with 0, there should be
an interference pattern, but when becomes the order of
c , the amplitude packets no longer fully overlap at the interferometer output and the pattern fades away.

dk i k s ,k i k s k i k s k i k 0 ,
13

where (k s k i k 0 ) is the Dirac delta function. The details

of this general state reside in the form of the amplitude
(k s ,k i ), which can be and usually is a function of many
more variables than k. If (k s ,k i ) is not factorable into
separate functions of k s and k i , the state is entangled. As we
have noted, the apparatus can shape (k s ,k i ) to be nonzero
over a restricted range of k around k 0 and result in an
amplitude packet arising from a coherent superposition of
correlated energy eigenstates with some range of energies
Eck and with a corresponding spatial extent.
In our experiments the bandwidth of a packet reaching our
detectors is determined chiefly by irises that define the paths

D. One-photon interference with different coherence

lengths
We control the value of c in our experiments by limiting
the spread of wave numbers k with bandpass filters placed
in front of the detectors. The properties of such filters are
usually expressed in terms of the spread of wavelengths
that they transmit around a central wavelength 0 . Hence, it
is customary to express c ct in terms of these quantities.
Because k 0 2 / 0 , it follows that k2 / 20 if is
small compared to 0 . Then the coherence length c
20 /. The data shown in Fig. 3 were taken with
915.8 nm down-conversion photons and with 10 nm bandpass filters in front of the detectors. The corresponding coherence length is 84 m.
As increased and approached c , the visibility of the
data in Fig. 3a decreased. For example, when we increased
to 36 m by turning the micrometer of the linear stage
where the piezo-driven mirror was mounted, we measured
fringes not shown with a lower visibility V0.310.05.
When was increased to 144 m, we obtained the data
shown by the circles in Fig. 3b. In other words, when the
path lengths through the interferometer differed by more than
c , the fringes disappeared into the noise. When the differ-

132
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
132
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

ence in path length through the interferometer arms is this

large, the packets do not overlap, and the apparatus is capable of distinguishing which path a given photon took. The
amplitudes then combine according to Feynmans rule 3, and
there is no interference, that is, P2RT.
As a dramatic demonstration of the effect of the magnitude of the coherence length, we placed a 0.1 nm filter in
front of the idler detector with 144 m. The new filter
increased the coherence length to 8400 m, and, as can be
seen by the squares in Fig. 3b, the fringes reappeared with
a visibility of V0.590.04.
Note that we placed the 0.1 nm filter in front of the idler
detector and not in front of the signal detector. Because of
the entanglement between the idler and the signal photons, it
makes no difference whether we change the filter in front of
the idler detector or the one in front of the signal detector.
The signal detector had a 10 nm filter in front of it. This
result is a dramatic demonstration of the energy correlation
between the two photons. Reference 20 gives an interesting
interpretation of this result in terms of the collapse of the
wavefunction of the idler and signal photons.

Fig. 4. Data for the interference of two collinear photons going through the
Michelson interferometer. The coincidence data correspond to the cases
when the path-length difference was approximately 0 squares, 47 m
circles, and 210 m triangles. The crosses () are the data collected by
a single detector when 37 m.

There are two interference patterns present, and the data

show a progression in which one pattern remains and the
other fades away as is increased.23

IV. BIPHOTON INTERFERENCE

B. Biphoton: Interpreting the results

In the one-photon experiment, interference arose from the

superposition of the two amplitudes describing two indistinguishable alternate paths by which a single photon could
produce the same outcome. If we send two collinear photons
into the interferometer, the number of alternatives increases
to four, and the outputs are determined by the superposition
of four amplitudes. We should expect that the interference
pattern arising from four amplitudes will be more complicated than the one observed in the one-photon experiment, as
is indeed the case. Two-photon experiments exhibit interference that is fourth-order in the electric field rather than
second-order as in the more familiar single-photon interference experiments described in Sec. III A.18,19
A. Biphoton: Experimental results

The pattern for 0 can be predicted by the same

method that we used to predict the one-photon interference
pattern. We identify the possible paths through the interferometer and write down the corresponding plane-wave amplitudes modified by appropriate factors of r and t as they pass
through the interferometer. Because there are two photons,
each path will have associated with it the product of two
single-photon amplitudes. The wave numbers in these factors
must be modified to take entanglement into account.
For example, consider the possibility corresponding to
case A of Fig. 5. Here the two entangled photons are viewed
as traveling together in arm 1. With r and t being respectively the reflection and transmission amplitudes of the interferometers input beam splitter, the amplitude is
rte i(k 0 k) 1 rte i(k 0 k) 1 r 2 t 2 e i2k 0 1 . The use of k in one

We did two-photon experiments with a Michelson interferometer used as described in Refs. 21 and 22. As shown in
Fig. 1b, photons from the 402.4 nm pump laser entered a
down-converter crystal oriented to emit a collinear pair of
equal-wavelength photons. The crystal sat between two
crossed polarizers. The first polarizer ensured that pump photons entered the crystal horizontally polarized, and the second polarizer ensured that only vertically polarized downconversion photons entered the interferometer. The entangled
pair of down-conversion photons passed through the Michelson interferometer and then to a beam splitter, each output
of which was viewed by an avalanche photodiode detector.
We recorded only the events that produced coincidences between the two detectors.
Our data are shown in Fig. 4. They are coincidences plotted as a function of , which was extracted from the fits to
the data, as described earlier. The data were taken with three
different interferometer settings: 0 squares, as verified
with white-light fringes, 47 m circles, and
210 m triangles. Both detectors had 10 nm filters in
front of them. As is particularly apparent in the data of solid
circles, the resulting interference patterns are the sum of two
oscillations, one with a frequency twice that of the other.

Fig. 5. Four possible paths for two collinear photons in a Michelson interferometer. The labeled components are nonpolarizing 50-50 beam splitter
B, mirror M, and mirror moved by a piezoelectric stack Mp. The output
ports are labeled a and b.

133
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
133
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

factor and k in the other takes into account entanglement

by specifying that if one photon differs from the central wave
number k 0 by k, the other must differ by k. For this
particular amplitude the two deviations from the central
wave number cancel.
For the path corresponding to two photons traveling
through arm 2, a similar argument gives t 2 r 2 e i2k 0 2 . Note
that the presence of two photons in the same arm of the
interferometer multiplies the phase of the amplitude by a
factor of 2.
We also must include the amplitudes for cases C and D of
Fig. 5. In case C, one photon, say the idler, is imagined as
traveling through arm 1 while the signal photon travels
through arm 2. The corresponding amplitude is
rte i(k 0 k) 1 tre i(k 0 k) 2 . There is a similar amplitude for the
same paths but with the two photons interchanged. In these
amplitudes there is no cancellation of the wave vector deviations k.
The probability of passing through the interferometer is
the square of the modulus of the sum of the probability amplitudes for cases A, B, C, and D,
P 2 r 2 t 2 e i2k 0 1 e i2k 0 2 2
cos k 1 2 e ik 0 ( 1 2 ) 2

14a

2R T 12 cos k 4
2 2

cos k cos cos 2 .

14b

Notice that if k1 Eq. 14 simplifies to

P 2 2R 2 T 2 3cos 2 4 cos

15a

4R 2 T 2 12 cos cos2

15b

4R T 1cos .

15c

2 2

The result of Eq. 15c predicts fringes that are narrower than
the single-photon interference fringes. In this respect these
quantum multiple-path interferences are analogous to multiplebeam wave interference: the more interfering paths or
beams, the narrower the fringes.24 The curve passing
through the squares in Fig. 4 shows that Eq. 15c provides
an excellent fit to our data for the case when 0.
We can arrive at the same theoretical interpretation using
the state vector formalism. We start with a symmetrized
wavefunction like Eq. 12, and replace the input wavefunctions by coherent superpositions of the wavefunctions for
going through each arm, as done in Sec. III A for the case of
the single photon. The photon leaving the interferometer
away from the source is in state a, and the photon returning
to the source is in state b. The state vectors then become
k0 k s,ire i(k 0 k) 1 t a s,ir b s,i)te i(k 0 k) 2 r a s,i

t b s,i).

16

&r 2 t 2 e i2k 0 1 e i2k 0 2 2e ik 0 ( 1 2 )cos k( 1 2 ) .

19
We substitute Eq. 19 into Eq. 18 to obtain Eqs. 14a and
14b.25
In our setup cases C and D can become distinguishable
while A and B remain indistinguishable. This case occurs
when we increase to about 210 m. Then we obtain the
data represented by the triangles in Fig. 4. If we apply Feynmans rule to a situation where cases C and D are distinguishable, we must add their contributions separately to the
final probability. Cases A and B remain indistinguishable, so
P 2 r 2 t 2 e i2k 0 1 e i2k 0 2 2 R 2 T 2 R 2 T 2

20a

2R 2 T 2 1cos 2 2R 2 T 2

20b

1
4R 2 T 2 1 cos 2 .
2

20c

Equation 20c predicts both the reduction in the magnitude of the amplitude and the doubling of the fringe frequency shown by the triangles in Fig. 4. For comparison note
that the single-photon fringes the crosses in Fig. 4 obtained
by one of the detectors when 37 m clearly have a
frequency which is half that of the two-photon fringes.
We fit our data with a version of Eq. 14b parameterized
to describe the visibilities of different parts of the overall
pattern:
N c N 0 2V 0 4V CD cos V AB cos 2 ,

21

where N c is the number of counts recorded in some chosen

time interval. V 0 and V CD are related to the visibility of the
one-photon pattern, and V AB/2 is the visibility of the twophoton interference pattern. We will give further justification
for this parameterization in the following section.
The data recorded when the difference in path lengths
through the interferometer is zero squares in Fig. 4 are well
fit by the solid line which is Eq. 21 with N 0 442
13 counts/(2s), V AB0.980.08, and V CD0.980.02
(V 0 0.990.02). That is, the data are consistent with Eq.
15; there is full indistinguishability of the four possibilities.
When the path lengths through the arms were made to differ
by 47 m, we obtained the data represented by the circles;
these are well fit using V AB0.810.09 and V CD0.15
0.07. These visibilities show that cases C and D are becoming distinguishable while A and B remain indistinguishable. Finally, when the path-length difference was increased
to 210 m, the part of the pattern arising from the possibilities of the photons taking separate paths that is, cases C and
D) essentially vanishes, and the data triangles in Fig. 4 are
fit using just V AB0.510.13 and V CD0.010.05. This
result is consistent with Eq. 20c.

After some algebra we find that the state vector is given by

a s a i a s b i b s a i b s b i ,

17

where , , , and are complex functions of the interferometer lengths and wavenumbers. Because we detect the
case when both photons leave the interferometer in state a,
the probability that both photons go through the interferometer is
P 2 a s a i 2 * ,
where

18

C. Biphoton: Calculating the coherence length

To understand why the above data behave as observed, we
need probability amplitudes that take into account the finite
spread in wave numbers. We avoid the complete theory,26 but
construct model amplitudes complicated enough to contain
the essential features of the experiment and simple enough to
allow students to work quantitatively with the concepts of
coherence length and entanglement. We make such model
amplitudes by assuming that the values of k that appear in

134
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
134
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

Eq. 14b are distributed uniformly over an interval k centered on k 0 . We then build the amplitudes by integrating the
relevant terms.
We replace 2 cos2(k) in Eq. 14b by (1cos 2k) and
integrate over the uniform distribution of k to find

P 2 2R 2 T 2 1
4

1
k

1
k

k/2

k/2

k/2

k/2

1cos 2k dk

cos k dkt cos cos 2 .

22

The first integral in Eq. 22 gives 1sinc(k) , and

the second gives sinc(k/2), where the sinc function is
defined as,
sinc x

sin x
,
x

23
V. POLARIZATION AND INTERFERENCE

and might be familiar to students who have analyzed singleslit diffraction. For x0, sinc(0)1, and as x increases,
sinc(x) oscillates with diminishing amplitude around the x
axis. Its first zero occurs when x , that is, sinc( )0.
The result is that with a uniform distribution of wave numbers, Eq. 14b predicts that
P 2 2R 2 T 2 2sinc k 4sinc k/2
cos cos 2 .

Fig. 6. Coincidences per second as a function of the orientation of polarizer

P 2 in front of the signal detector for two settings of polarizer P 1 located as
shown schematically in the insert.

24

The sinc terms result in the loss of interference when the

distance a photon travels in one arm of the interferometer
becomes appreciably different from the distance it travels in
the other. Both sinc functions go to zero when becomes
appreciably larger than 1/k. The smallest value of the argument for which both become identically zero is k
2 . This value occurs when 2 /k, the quantity
earlier identified as the coherence length c .
When either k0 as for our earlier plane wave treatment or when 0 white-light fringes, Eq. 24 reduces
to Eq. 15a. And when the difference in the lengths of the
paths that they travel through the interferometer arms becomes large relative to the coherence length, the terms in Eq.
24 multiplied by the sinc functions drop out, and Eq. 24
reduces to Eq. 20c.
This analysis confirms our earlier interpretation. The interference patterns of Fig. 4 arise from two different modes of
interference. In the AB mode the biphoton produces the interference pattern of a single photon with the wave number
and the longer coherence length of the pump laserso long
that we did not include it in our analysis. The CD mode of
the biphoton produces single-photon interference of downconversion photons with the much shorter coherence length
that they acquire in down conversion. The presence of the
two modes and the fading of the CD mode as is increased
to and beyond the CD coherence length explain the progression of the curves in Fig. 4.
Our biphoton experiments are a reminder to be careful
interpreting Diracs statement that interference between two
different photons never occurs. 17 As Glauber has pointed
out,27 it is amplitudes that interfere. The biphoton illustrates
that interference can occur regardless of the number of photons involved.

A. Polarizer as a wavefunction projector

We can use a polarizer to transform the polarization states
of our photons. The effect of the polarizer is conveniently
described in terms of vectors corresponding to its transmission and extinction directions. If we represent the horizontally and vertically polarized states of photons as H and
V , respectively, the polarizer with its transmission axis
forming an angle with the horizontal will transform the
photon states into the new basis states
T cos H sin V

25

for the transmission axis, and

E sin H cos V

26

for the extinction axis. Because the polarizer transmits photons in the T state and absorbs photons in the E state,
the transmission probability of a photon in state is
P T 2 .

27

As we have noted, the type-I down conversion of our experiments generates two vertically polarized photons, that is,
in the state V. In an experiment where the down-conversion
beams went directly to the detectors, we placed two polarizers, labeled 1 and 2, on the path of the signal beam as shown
in the insert to Fig. 6. We set polarizer 1 with its transmission
axis vertical that is, 1 /2), and varied the angle 2 of the
second polarizer. The circles in Fig. 6a show that the variation of the coincidences as a function of 2 is well described
by
P T 2 V 2 sin2 2 .

28

But the polarizer does much more: it also projects the

photon state onto the transmission state. That is, the state of
the photon after the polarizer is T T . If we rotate
polarizer 1 so that it forms an angle 1 with the horizontal,
then the state of the photon before the second polarizer will
be T 1 T 1 V , and the transmission probability will be
P T 2 T 1 T 1 V 2 cos2 2 1 sin2 1 .

29

The squares in Fig. 6 represent the data for 1 3 /4, and

are consistent with

135
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
135
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

P 21 cos2 2 3 /4 .

30

The lines in Fig. 6 are fits of the function N c N 0 cos2(2

) to the two data sets. The amplitudes N 0 of the fits have
a ratio of 0.450.10, and the fitted phases differ by 41
3, consistent with our expectations. Notice that the
minima of the two graphs in Fig. 6 do not occur at the same
angles. When 1 /2,3 /2, the probability is no longer
zero at 2 n integer n).
B. The quantum eraser
Our quantum eraser experiment, which is like those of
Ref. 28, further illustrates the importance of indistinguishability and which-way information. Schneider and
LaPuma29 have recently reported using an attenuated source
to do a similar experiment.
We arranged our apparatus as described in Sec. III A, but
with a first-order quartz half-wave plate in one arm of the
interferometer and a dummy quartz wave plate in the other
arm to compensate for the added optical path length dashed
rectangles in Fig. 1. By aligning either axis of the half-wave
plate fast or slow with the polarization of the downconversion light that is, vertical and stepping the voltage v p
on the piezo, we obtained the graph shown by the squares in
Fig. 7. The result is well described by a parameterized version of Eq. 8. A least-squares fit to N 0 (1V cos ) gave
N 0 39220 counts per 20 s and V0.790.06.
The rotation of the half-wave plate by an angle rotates
the polarization by 2. The state of the photon emerging
from arm 1 is given by
1 rt cos 2 V sin 2 H ],

VI. DOES THE PHOTON EXIST?

PRT e i 1 1 e i 2 V 2 21 1cos 2 cos , 32

for RT1/2. When /4, there is no interference that
is, P1/2) because the apparatus associates a distinct polarization state with each possible path: V for one path and
H for the other. This setting of the wave plate has the effect
of making the paths distinguishable, and the presence of such
distinguishing information results in the absence of interference even if the photons polarization state is not measured.
The circles in Fig. 7 show the data obtained when /4. A
fit gives N 0 28618 (20 s) 1 and V0.010.08, showing that there is indeed no interference.
We can erase the distinguishing information by projecting
the two orthogonal states with a polarizer set to /4 relative
to the horizontal and located after the interferometer. The
projection amplitudes between H and V and the polarizer
state T /4 are the same. The state of each photon after such
a polarizer is
33

so the probability of detecting a photon is

P 21 1cos .

0.01 at these wavelengths. If we correct for this attenuation, the ratio of the two amplitudes is 0.470.10, consistent
with the predicted 1/2.
We have introduced this experiment with /4) in our
first-year introductory physics class. The experiment illustrates some of the basic ideas of quantum mechanics discussed in this course.30 Because there was only one setup,
students took turns doing the experiments. The results were
explained using the concepts of the distinguishability of
paths.

31

and the probability of detecting it is

T /4 T /4 H e i 1 T /4 V e i 2 ,

Fig. 7. Data from the quantum eraser experiment when the interferometer
paths are indistinguishable squares, distinguishable circles, and when the
distinguishing information is erased by placing a polarizer after the interferometer triangles.

34

The triangles in Fig. 7 show the data obtained with a polarizer oriented at /4 after the interferometer. A fit to the
data gives N 0 648 (20 s) 1 and V0.820.12, showing
that the polarizer erased the distinguishing information. The
inserted polarizer inherently attenuated the light by 0.46

Is there actually such a thing as a photon? We wanted

students to confront this question, because it is fundamental
to the interpretation of our experiments and because the evidence for the photon is less obvious than students are given
to think. The Compton effect and the photoelectric effect are
usually cited as evidence for the photons existence, but both
effects can be explained as arising from the interaction of a
continuous classical electromagnetic field with matter possessing quantized energy states. Such semiclassical theories have no photons; quantum properties are associated only
with matter, not with the electromagnetic field.
Tagging the signal photon with the idler photon allows us
to say with considerable certainty that the amount of energy
in our apparatus corresponds to the energy of a single photon. Consequently, if we direct this energy onto a 50-50
beam splitter and look with detectors at the two outputs, we
should be able to distinguish experimentally between the
semiclassical theory with its continuous field and the fully
quantum theory with its discrete photons.
In the semiclassical case the two detectors should register
counts in coincidence with each other. The semiclassical explanation of the photoelectric effect31 predicts that detectors
using the photoelectric effect and sensitive to some frequency f the kind of detector used in most quantum optics
experiments, including ourscan register counts when illuminated with light down to arbitrarily low intensity levels. If
there were no such thing as a photon, an electromagnetic
wave striking a 50-50 beam splitter would divide if the semiclassical wave picture is applicable. Half of its energy would

136
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
136
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

Fig. 8. Interference data for signal photons that go through a MachZehnder

interferometer and reach a beam splitter with two detectors S A and S B located at its output ports as shown in the insert. The circles and squares
correspond to double coincidences of the idler detector with S A and S B ,
respectively. Symbols () correspond to triple coincidences of the idler
detector, S A , and S B .

emerge from one output and half from the other, and the
detector at one output should detect electromagnetic energy
in coincidence with the detector at the other.
If the field is quantized, however, there should be no coincidences. If the field quantum, the photon, exists and is
indivisible, and if there is only one in the system, the detection of a photon at one output means no photon will be
detected at the other. In the absence of background radiation,
we should never detect electromagnetic radiation at one detector in coincidence with the other.
With our apparatus we could have performed a version of
the 1986 experiment by Grangier, Roger, and Aspect,32
which shows conclusively that photons never divide at a
beam splitter. Greenstein and Zajonc31 have given a particularly clear and well written analysis of the results of Ref. 32.
Recently, Beck and colleagues have used down-conversion
to do a similar experiment in an undergraduate setting.33
We did not test photon behavior at a beam splitter as thoroughly as in Refs. 32 or 33. Instead, we did an experiment
that could be set up quickly from the existing configuration
for the interference measurements and give results in a few
minutes of data taking. Our approach was to add a third
beam splitter at the output of the interferometer, with a lensfilter-detector set located at each of its output ports, as shown
in the insert in Fig. 8. The signal photons could then either
transmit to detector S A , reflect to detector S B , or potentially
split and be detected by both. For every signal photon passing through the beam splitter that is, via transmission or
reflection, the probability of detecting it is P A P B pbs ,
where pbs is the detection efficiency past the beam splitter.
For our apparatus we estimated pbs(915.8 nm)0.18
0.02, the product of the transmission efficiency through
the lens and filter (0.510.02) and the quantum efficiency of
the detector (0.360.02). 6 The probability for two simulta2
neous detections at detectors A and B is P AB pbs
pbsP A pbsP B . Thus, if every photon that contributes to
the interference pattern splits at the third beam splitter, the
triple coincidences between the idler, S A and S B should be
about 18% of the coincidences between the idler and S A or
SB .
We used a second TAC/SCA and another coincidence cir-

cuit to detect double coincidences of the idler detector with

S A and the idler detector with S B ; triple coincidences were
recorded as coincidences of the doubles. The idler detector
had a 1 nm filter in front of it, and the two signal detectors
had 10 nm filters in front of them.
The time window for coincidences in the doubles was set
to 40 ns. Each double coincidence was recorded when the
signal photon A or B arrived within 10 ns of the idler
photon. The predicted rate of accidental coincidence of
doubles for our setup was 2106 s1 . We tested our
coincidence electronics with fake triple coincidence pulses
generated by a digital-delay pulse generator.
Figure 8 shows three sets of data. The coincidences between the idler detector and signal detector S A squares, and
the coincidences between the idler detector and signal detector S B circles show interference fringes as the path length
difference of the interferometer is varied via the piezo voltage v p . The triple coincidences ( in Fig. 8 do not show
any counts. As we noted, had each of the photons split at the
third beam splitter, we would expect an interference pattern
in the triple coincidences with an amplitude of about 12
counts in 10 s. Consequently, our data are consistent with the
conclusion that some appreciable number of the photons do
not split.
VII. SUMMARY AND CONCLUSIONS
We have developed undergraduate level experiments that
use a source of correlated pairs of photons to illustrate basic
principles of quantum mechanics. The results from our tabletop experiments are simply explained when the interferometers are adjusted to have nearly equal length arms. In this
case we can use Feynmans explanation of quantum interference in terms of simple plane-wave probability amplitudes.
With the help of the concept of coherence length, more complicated situations of interference can be understood in terms
of distinguishability and indistinguishability. The interference of single photons also provides a physical situation in
which students can learn to use state vectors to investigate
state projection, basis change, and the calculation of probabilities for a particular outcome. The experiments also provide dramatic teaching moments for the discussion of fundamental questions about the nature of light and the concepts
of quantum mechanics.
The experimental setup fits on a 2 4 optical breadboard, requiring laboratory components that individually do
not exceed $7000. The experiments can be done without previous research experience with photon quantum optics.
The single-photon experiments presented here can be used
in undergraduate courses in quantum mechanics for which
laboratories and demonstrations are rare. By replacing the
down conversion crystal, the apparatus can be used to measure the violation of Bells inequalities.2 The building phase
of each experiment makes an excellent upper level undergraduate project.
ACKNOWLEDGMENTS
We are indebted to P. Kwiat, whose ideas and help were
invaluable to the success of this project. We also thank A.
Zeilinger and T. Jennewein for their assistance, M. Beck, J.
Eberly, S. Malin, M. E. Parks, and W. Wooters for helpful
discussions, and S. J. Hinterlong and R. Williams for help

137
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
137
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

Table I. Prices of essential components of the basic setup.

Description

Vendor

Model

Each

No.

Comment
402.3 nm, 18 mW
beta-barium-borate,
775 mm
prism mount and rotation
stage
avalanche photodiode
10 nm filter for 2 p
XY translation
module
computer card
4 counter PC card
2 4

Diode laser
Crystal

Power Technology
Cleveland Crystals

1Q2C18/5911

$6400
$1000

1
1

Crystal mounting

Thorlabs

PR01KMPM

$370

Detectors
Filters
Lens mount
TAC/SCA
Multichannel scaler
Counter
Optical breadboard

EG&G/Pacer
Andover
Thorlabs
Canberra
Canberra
National Instruments
Thorlabs

SPCM-AQR-13
CW/L
LM1XY
2145
ASA-100
PCI6601BNC2121
T2448A

$4100
$600
$140
$1840
$3500
$700
$1150

2
2
2
1
1
1
1

with the equipment. This work was funded by a grant from

the National Science Foundation DUE-9952626.

hardware, where the optics are very close to the optical table,
to be less versatile but simpler, less expensive, and more
stable against vibrations.

APPENDIX A: COST OF COMPONENTS

APPENDIX B: DOWN CONVERSION

We did not have prior experience with these types of experiments, and the cost of the essential components listed in
Table I is intended for those with similar inexperience. The
total cost of this essential equipment is about $25,000. Table
I does not include the standard mounting hardware or the
optical hardware to steer the laser onto the down-conversion
crystal, which may add about $1,000 to the total price. The
price of the blue diode laser that we list is for a module that
includes current and temperature control and beam-shaping
optics. The price of the bare laser diode is much lower
$10002000.2
The essential equipment items for the experiments that we
discussed are listed in Table II. The total cost of these elements is about $6000. It does not include the mounting hardware and apparatus for steering the down-conversion beams
in and out of the interferometer 4 mirrors plus mounting
hardware. If we include all of them, the cost is about $9000.
A picture of the apparatus is shown in Fig. 9. As stated, the
estimated cost is for plug and play parts, provided all is
put together carefully. The cost can be lowered by customizing the electronics and hardware parts. We also tried several
hardware arrangements. Most experiments were done with
optics mounted in convenient magnetic mounts, linked by
rods in the interferometers. We found pedestal mounting

Spontaneous parametric down conversion is a well studied

nonlinear optics effect. In brief, it involves the use of a birefringent crystal to convert an incident pump photon into two
photons, the signal and the idler. As mentioned, this process
conserves energy Eck, so that
k p k s k i ,

B1

where k p , k s , and k i are the wave numbers in vacuum. Here

we treat only degenerate down conversion where k s k i
k p /2. Consider the case where the down-conversion photons leave the crystal in different directions as shown in Fig.
1a. If c is the angle that the signal and idler photons form
with the direction of propagation of the pump beam inside
the crystal, then Eq. 3 becomes
n p k p 2n s k s cos c .

B2

We combine k s k p /2 with Eq. B2 and obtain

n p n s cos c .

B3

It is not possible to satisfy Eq. B3 in an isotropic medium,

because for normal dispersion the index of refraction decreases with increasing wavelength, that is, n p n s . This
problem can be overcome with a birefringent crystal. The top

Table II. Prices of essential components of the experiments.

Description
Cube beam splitter
Prism mount
Mirror mount
Piezo stack
Linear stage
Filter
Wave plate
Wave plate
Polarizer
Voltage amplifier
PC interface

Vendor

Model

Each

No.

Comment

Melles Griot
New Focus
Thorlabs
Thorlabs
Thorlabs
Andover
Melles Griot
Melles Griot
Edmund
Trek
National Instruments

03BSC027
9411
KS1
AE0505D8
MT1
CW/L
02WRQ0023
02WRQ0003
A46088
P0516A-1
PCI6703
BNC2121

$120
$310
$80
$130
$250
$850
$690
$590
$250
$700
$1400

2
2
2
1
1
1
1
1
2
1
1

near-IR, nonpolarizing
Beam-splitter mount
high-stability mounts
for changing
piezo mount goes on top
1 nm filter for 2 p
half wave
quarter wave
Near-IR polarizer
0150 V to drive piezo
PC card, connector box
and Labview software

138
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
138
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

dex of refraction n e ( m) and the down-conversion photons

have the ordinary index of refraction. Suppose that we wish
to have the two down-conversion photons collinear, that is,
c 0 in Eq. B3. Then by setting m25.67, the pump
beam will have the index of refraction given by the dashed
e ( m)1.658 at 457.9 nm. If we
line in Fig. 10, or n p n
want the signal and idler beams to form a laboratory angle of
L3 with the pump beam outside the crystal, we can use
Snells law, sin Ln s sin c , to obtain c and find the phase
matching angle m that satisfies Eq. B3.
1

Fig. 9. Photograph of the layout for the biphoton experiments. The hardware
is mounted on pedestal mounts, and the entire layout fits on a 2 4 optical
breadboard. The path of the light beams is traced in white.

solid line in Fig. 10 represents the ordinary index of refraction, where the polarization of the light is perpendicular to
the optic axis OA of the crystal. If the polarization is in the
same plane as OA, the index of refraction, also known as the
extraordinary index of refraction, depends on the angle m
formed between the propagation direction kp and OA. The
lower solid line in Fig. 10 corresponds to the case where
e ( m /2), with
n e n
n e m cos2 m /n 2o sin2 m /n 2e 1/2.

B4

By means of m we tune the extraordinary index of

refraction between n o and n e . The graphs of the indices of
refraction shown in Fig. 10 correspond to those of the negative uniaxial beta-barium-borate crystal, with the index of
refraction given by34

n A

B
D 2
2
C

1/2

B5

where the constants for n o and n e are A o 2.7359, B o

0.01878 m2 , C o 0.01822 m2 , D o 0.01354
m2
and
A e 2.3753,
B e 0.01224 m2 ,
Ce
2
0.01667 m , and D e 0.01516 m2 . For betabarium-borate we can verify that n o 1.658 at 915.8 nm and
n e 1.542 at 457.9 nm. Under the situation known as type-I
phase matching, the pump photon has the extraordinary in-

Fig. 10. Index of refraction curves for a beta-barium-borate crystal with its
optic axis aligned perpendicular top and parallel bottom to the input
polarization. The dashed middle curve corresponds to the phase-matching
condition that allows down-conversion for the setup of Fig. 1b.

The Physics of Quantum Information, edited by D. Bouwmeester, A. Ekert,

and A. Zeilinger Springer, Berlin, 2000.
2
D. Dehlinger and M. W. Mitchell, Entangled photon apparatus for the
undergraduate laboratory, Am. J. Phys. 70, 898 902 2002; ibid., Entangled photons, nonlocality, and Bell inequalities in the undergraduate
laboratory, Am. J. Phys. 70, 903910 2002.
3
C. H. Holbrow, E. J. Galvez, and M. E. Parks, Photon quantum mechanics and beam splitters, Am. J. Phys. 70, 260265 2002.
4
Richard P. Feyman, QED, The Strange Theory of Light and Matter Princeton University Press, Princeton, 1985.
5
S. Friedberg, C. K. Hong, and L. Mandel, Measurement of time delays in
the parametric production of photon pairs, Phys. Rev. Lett. 54, 2011
2013 1985.
6
Single Photon Counting Module SPCM-AQR Series, Perkin Elmer Optoelectronics product report Perkin Elmer, Vaudreuil, 2001.
7
Since we completed this work, the available power levels for the 400 410
nm diode lasers have risen to 50 mW. In addition, new diode lasers at 375
nm have become available.
8
If the initial propagation direction is 1,0,0, with polarization along
0,0,1, then a sequence of mirrors that send the beam consecutively
through the directions 0,0,1, 0,1,0, 1,0,0 rotates the polarization by
90. Omitting the last step allows the same rotation when steering the
beam into the 0,1,0 direction.
9
For more information, see http://departments.colgate.edu/physics/
pql.htm.
10
R. P. Feynman, R. B. Leighton, and M. Sands, The Feynman Lectures on
Physics Addison-Wesley, Reading, 1965, Vol. 3, p. 1-1.
11
Note that e ik is an example of Feynmans formulation of quantum mechanics in terms of the action: e i pdx/ . See R. P. Feynman, The Theory of
Fundamental Processes W. A. Benjamin, New York, 1962.
12
We use j nk j j k j j because n1 for experiments done in air.
13
D. M. Greenberger, M. A. Horne, and A. Zeilinger, Multiparticle interferometry and the superposition principle, Phys. Today 468, 2229
1993.
14
In this case the /2 relative phase between r and t yields P( )2RT 1
cos .
15
The pattern of the maxima and minima will be the same for tagged photons as for untagged photons. A tagged photon, however, is a special case
of a biphoton which is discussed later, and its pattern of partial interference is not the same as that of an untagged photon.
16
We fit the data using the solver option of Excel, minimizing the reduced
2 . The errors were calculated using the 2 1 criterion. K. E. Schalm,
Least-squares fits to an arbitrary function using Excel, handout for
PHY 445/PHY 515 Advanced Laboratory course at SUNY, Stony Brook,
9/18/95, unpublished.
17
P. A. M. Dirac, The Principles of Quantum Mechanics Clarendon Press,
Oxford, 1958, p. 9.
18
R. Ghosh, C. K. Hong, Z. Y. Ou, and L. Mandel, Interference of two
photons in parametric down conversion, Phys. Rev. A 34, 39623968
1986.
19
C. K. Hong, Z. Y. Ou, and L. Mandel, Measurement of subpicosecond
time intervals between two photons by interference, Phys. Rev. Lett. 59,
2044 2046 1987.
20
P. G. Kwiat and R. Y. Chiao, Observation of a nonclassical Berrys phase
for the photon, Phys. Rev. Lett. 66, 588 591 1991.
21
E. Mohler, J. Brendel, R. Lange, and W. Martienssen, Finesse and resolution enhancement in two-photon interferometry, Europhys. Lett. 8,
511516 1989.
22
J. Brendel, E. Mohler, and W. Martienssen, Time-resolved dual-beam
two-photon interferences with high visibility, Phys. Rev. Lett. 66, 1142
1145 1991.

139
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
139
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39

23

Due to the better alignment and simpler optical arrangement that we can
obtain in the experiment with collinear photons, the coincidence yield is
much higher than in the experiments with the noncollinear photons.
24
Notice that the result is just the square of the result for one-photon interference. In other words, the interference pattern of the biphoton is the
product of the interference patterns of single photons.
25
We detect coincidences at the interferometer output by placing a beam
splitter there with reflectance R and transmittance T and then detecting
coincidences between photons leaving each output. This arrangement allows us to detect only 2R T of the photon pairs leaving the interferometer.
26
R. A. Campos, B. E. A. Saleh, and M. A. Teich, Fourth-order interference
of joint single-photon wavepackets in lossless optical systems, Phys. Rev.
A 42, 4127 4137 1990.
27
Roy J. Glauber, Diracs famous dictum on interference: One photon or
two?, Am. J. Phys. 63, 12L 1990. We thank H. Leff for bringing this
reference to our attention.
28
P. D. D. Schwindt, P. G. Kwiat, and B.-G. Englert, Quantitative wave-

particle duality and nonerasing quantum erasure, Phys. Rev. A 60, 4285
4290 1999.
29
M. B. Schneider and I. A. LaPuma, A simple experiment for discussion
of quantum interference and which-way measurement, Am. J. Phys. 70,
266 271 2002.
30
C. H. Holbrow, J. N. Lloyd, and J. C. Amato, Modern Introductory Physics
Springer-Verlag, New York, 1999.
31
G. Greenstein and A. G. Zajonc, The Quantum Challenge Jones and Bartlett, Sudbury, 1987.
32
P. Grangier, G. Roger, and A. Aspect, Experimental evidence for a photon anticorrelation effect on a beam splitter: A new light on single-photon
interferences, Europhys. Lett. 1, 173179 1986.
33
J. J. Thorn, M. S. Neel, V. W. Donato, G. S. Bergreen, R. E. Davies, and
M. Beck, Observing the quantum behavior of light in an undergraduate
laboratory, Am. J. Phys. 72, 12101219 2004.
34
K. Kato, Second harmonic generation of 2048 -BaB2 O4 , IEEE J.
Quantum Electron. QE-22, 10131014 1986.

NEWTONS BEQUEST
So gravity was not mechanical, not occult, not a hypothesis. He had provided it by mathematics. It is enough, he said, that gravity really exists and acts according to the laws that we have
set forth and is sufficient to explain all the motions of the heavenly bodies and of our sea. It could
not be denied, even if its essence could not be understood.
He had declared at the outset that his mission was to discover the forces of nature. He deduced
forces from celestial bodies motion, as observed and recorded. He made a great claimthe
System of the Worldand yet declared his program incomplete. In fact, incompleteness was its
greatest virtue. He bequeathed to science, that institution in its throes of birth, a research program,
practical and open-ended. There was work to do, predictions to be computed and then verified.
James Gleick, Isaac Newton Vintage Books, 2003, pp. 139140.

140
Am. J. Phys., Vol. 73, No. 2, February 2005
Galvez et al.
140
This article is copyrighted as indicated in the article. Reuse of AAPT content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
168.176.55.14 On: Wed, 08 Apr 2015 20:56:39